library(matrixStats) # for numerically stable logsumexps options(amelia.parallel="no", amelia.ncpus=1) library(Amelia) source("pl_methods.R") source("measerr_methods_2.R") ## for my more generic function. run_simulation_depvar <- function(df, result, outcome_formula = y ~ x + z, coder_formulas = list(y.obs.0 ~ 1, y.obs.1 ~ 1), proxy_formula = w_pred ~ y.obs.1+y.obs.0+y){ (accuracy <- df[,mean(w_pred==y)]) result <- append(result, list(accuracy=accuracy)) (error.cor.z <- cor(df$x, df$w_pred - df$z)) (error.cor.x <- cor(df$x, df$w_pred - df$y)) (error.cor.y <- cor(df$y, df$y - df$w_pred)) result <- append(result, list(error.cor.x = error.cor.x, error.cor.z = error.cor.z, error.cor.y = error.cor.y)) model.null <- glm(y~1, data=df,family=binomial(link='logit')) (model.true <- glm(y ~ x + z, data=df,family=binomial(link='logit'))) (lik.ratio <- exp(logLik(model.true) - logLik(model.null))) true.ci.Bxy <- confint(model.true)['x',] true.ci.Bzy <- confint(model.true)['z',] result <- append(result, list(lik.ratio=lik.ratio)) result <- append(result, list(Bxy.est.true=coef(model.true)['x'], Bzy.est.true=coef(model.true)['z'], Bxy.ci.upper.true = true.ci.Bxy[2], Bxy.ci.lower.true = true.ci.Bxy[1], Bzy.ci.upper.true = true.ci.Bzy[2], Bzy.ci.lower.true = true.ci.Bzy[1])) (model.naive <- lm(y~w_pred+z, data=df)) naive.ci.Bxy <- confint(model.naive)['w_pred',] naive.ci.Bzy <- confint(model.naive)['z',] result <- append(result, list(Bxy.est.naive=coef(model.naive)['w_pred'], Bzy.est.naive=coef(model.naive)['z'], Bxy.ci.upper.naive = naive.ci.Bxy[2], Bxy.ci.lower.naive = naive.ci.Bxy[1], Bzy.ci.upper.naive = naive.ci.Bzy[2], Bzy.ci.lower.naive = naive.ci.Bzy[1])) loa0.feasible <- glm(y.obs.0 ~ x + z, data = df[!(is.na(y.obs.0))], family=binomial(link='logit')) loa0.ci.Bxy <- confint(loa0.feasible)['x',] loa0.ci.Bzy <- confint(loa0.feasible)['z',] result <- append(result, list(Bxy.est.loa0.feasible=coef(loa0.feasible)['x'], Bzy.est.loa0.feasible=coef(loa0.feasible)['z'], Bxy.ci.upper.loa0.feasible = loa0.ci.Bxy[2], Bxy.ci.lower.loa0.feasible = loa0.ci.Bxy[1], Bzy.ci.upper.loa0.feasible = loa0.ci.Bzy[2], Bzy.ci.lower.loa0.feasible = loa0.ci.Bzy[1])) ## df.loa0.mle <- copy(df) ## df.loa0.mle[,y:=y.obs.0] ## loa0.mle <- measerr_mle_dv(df.loa0.mle, outcome_formula=outcome_formula, proxy_formula=proxy_formula) ## fisher.info <- solve(loa0.mle$hessian) ## coef <- loa0.mle$par ## ci.upper <- coef + sqrt(diag(fisher.info)) * 1.96 ## ci.lower <- coef - sqrt(diag(fisher.info)) * 1.96 ## result <- append(result, list(Bxy.est.loa0.mle=coef['x'], ## Bzy.est.loa0.mle=coef['z'], ## Bxy.ci.upper.loa0.mle = ci.upper['x'], ## Bxy.ci.lower.loa0.mle = ci.lower['x'], ## Bzy.ci.upper.loa0.mle = ci.upper['z'], ## Bzy.ci.lower.loa0.mle = ci.upper['z'])) loco.feasible <- glm(y.obs.0 ~ x + z, data = df[(!is.na(y.obs.0)) & (y.obs.1 == y.obs.0)], family=binomial(link='logit')) loco.feasible.ci.Bxy <- confint(loco.feasible)['x',] loco.feasible.ci.Bzy <- confint(loco.feasible)['z',] result <- append(result, list(Bxy.est.loco.feasible=coef(loco.feasible)['x'], Bzy.est.loco.feasible=coef(loco.feasible)['z'], Bxy.ci.upper.loco.feasible = loco.feasible.ci.Bxy[2], Bxy.ci.lower.loco.feasible = loco.feasible.ci.Bxy[1], Bzy.ci.upper.loco.feasible = loco.feasible.ci.Bzy[2], Bzy.ci.lower.loco.feasible = loco.feasible.ci.Bzy[1])) ## df.double.proxy <- copy(df) ## df.double.proxy <- df.double.proxy[,y.obs:=NA] ## df.double.proxy <- df.double.proxy[,y:=NA] ## double.proxy.mle <- measerr_irr_mle_dv(df.double.proxy, outcome_formula=y~x+z, outcome_family=binomial(link='logit'), coder_formulas=list(y.obs.0 ~ y), proxy_formula=w_pred ~ y.obs.0 + y, proxy_family=binomial(link='logit')) ## print(double.proxy.mle$hessian) ## fisher.info <- solve(double.proxy.mle$hessian) ## coef <- double.proxy.mle$par ## ci.upper <- coef + sqrt(diag(fisher.info)) * 1.96 ## ci.lower <- coef - sqrt(diag(fisher.info)) * 1.96 ## result <- append(result, list(Bxy.est.double.proxy=coef['x'], ## Bzy.est.double.proxy=coef['z'], ## Bxy.ci.upper.double.proxy = ci.upper['x'], ## Bxy.ci.lower.double.proxy = ci.lower['x'], ## Bzy.ci.upper.double.proxy = ci.upper['z'], ## Bzy.ci.lower.double.proxy = ci.lower['z'])) df.triple.proxy <- copy(df) df.triple.proxy <- df.triple.proxy[,y.obs:=NA] df.triple.proxy <- df.triple.proxy[,y:=NA] triple.proxy.mle <- measerr_irr_mle_dv(df.triple.proxy, outcome_formula=outcome_formula, outcome_family=binomial(link='logit'), coder_formulas=coder_formulas, proxy_formula=proxy_formula, proxy_family=binomial(link='logit')) print(triple.proxy.mle$hessian) fisher.info <- solve(triple.proxy.mle$hessian) print(fisher.info) coef <- triple.proxy.mle$par ci.upper <- coef + sqrt(diag(fisher.info)) * 1.96 ci.lower <- coef - sqrt(diag(fisher.info)) * 1.96 result <- append(result, list(Bxy.est.triple.proxy=coef['x'], Bzy.est.triple.proxy=coef['z'], Bxy.ci.upper.triple.proxy = ci.upper['x'], Bxy.ci.lower.triple.proxy = ci.lower['x'], Bzy.ci.upper.triple.proxy = ci.upper['z'], Bzy.ci.lower.triple.proxy = ci.lower['z'])) ## df.loco.mle <- copy(df) ## df.loco.mle[,y.obs:=NA] ## df.loco.mle[(y.obs.0)==(y.obs.1),y.obs:=y.obs.0] ## df.loco.mle[,y.true:=y] ## df.loco.mle[,y:=y.obs] ## print(df.loco.mle[!is.na(y.obs.1),mean(y.true==y,na.rm=TRUE)]) ## loco.mle <- measerr_mle_dv(df.loco.mle, outcome_formula=outcome_formula, proxy_formula=proxy_formula) ## fisher.info <- solve(loco.mle$hessian) ## coef <- loco.mle$par ## ci.upper <- coef + sqrt(diag(fisher.info)) * 1.96 ## ci.lower <- coef - sqrt(diag(fisher.info)) * 1.96 ## result <- append(result, list(Bxy.est.loco.mle=coef['x'], ## Bzy.est.loco.mle=coef['z'], ## Bxy.ci.upper.loco.mle = ci.upper['x'], ## Bxy.ci.lower.loco.mle = ci.lower['x'], ## Bzy.ci.upper.loco.mle = ci.upper['z'], ## Bzy.ci.lower.loco.mle = ci.lower['z'])) ## my implementatoin of liklihood based correction mod.zhang <- zhang.mle.dv(df.loco.mle) coef <- coef(mod.zhang) ci <- confint(mod.zhang,method='quad') result <- append(result, list(Bxy.est.zhang = coef['Bxy'], Bxy.ci.upper.zhang = ci['Bxy','97.5 %'], Bxy.ci.lower.zhang = ci['Bxy','2.5 %'], Bzy.est.zhang = coef['Bzy'], Bzy.ci.upper.zhang = ci['Bzy','97.5 %'], Bzy.ci.lower.zhang = ci['Bzy','2.5 %'])) print(df.loco.mle) # amelia says use normal distribution for binary variables. tryCatch({ amelia.out.k <- amelia(df.loco.mle, m=200, p2s=0, idvars=c('y','ystar','w','y.obs.1','y.obs.0','y.true')) mod.amelia.k <- zelig(y.obs~x+z, model='ls', data=amelia.out.k$imputations, cite=FALSE) (coefse <- combine_coef_se(mod.amelia.k, messages=FALSE)) est.x.mi <- coefse['x','Estimate'] est.x.se <- coefse['x','Std.Error'] result <- append(result, list(Bxy.est.amelia.full = est.x.mi, Bxy.ci.upper.amelia.full = est.x.mi + 1.96 * est.x.se, Bxy.ci.lower.amelia.full = est.x.mi - 1.96 * est.x.se )) est.z.mi <- coefse['z','Estimate'] est.z.se <- coefse['z','Std.Error'] result <- append(result, list(Bzy.est.amelia.full = est.z.mi, Bzy.ci.upper.amelia.full = est.z.mi + 1.96 * est.z.se, Bzy.ci.lower.amelia.full = est.z.mi - 1.96 * est.z.se )) }, error = function(e){ message("An error occurred:\n",e) result$error <- paste0(result$error,'\n', e) }) ## mle.irr <- measerr_irr_mle( df, outcome_formula = outcome_formula, rater_formula = rater_formula, proxy_formula=proxy_formula, truth_formula=truth_formula) ## fisher.info <- solve(mle.irr$hessian) ## coef <- mle.irr$par ## ci.upper <- coef + sqrt(diag(fisher.info)) * 1.96 ## ci.lower <- coef - sqrt(diag(fisher.info)) * 1.96 ## result <- append(result, ## list(Bxy.est.mle = coef['x'], ## Bxy.ci.upper.mle = ci.upper['x'], ## Bxy.ci.lower.mle = ci.lower['x'], ## Bzy.est.mle = coef['z'], ## Bzy.ci.upper.mle = ci.upper['z'], ## Bzy.ci.lower.mle = ci.lower['z'])) return(result) }